Improving Deep Learning Classifiers Performance using Preprocessing and Cycle Scheduling Approaches in a Plant Disease Detection

Abstract

The impact of plant diseases on the growth of their corresponding species enhances the critical importance of early identification. Farmers currently grapple with significant challenges posed by these diseases, jeopardizing entire seasonal yields. To address this pressing issue, this paper introduces a ResNet-based Convolutional Neural Network model designed to detect various early-stage plant diseases. The dataset employed comprises approximately 46,000 RGB images of both healthy and diseased crop leaves, categorized into six distinct classes sourced from Plant Village. The dataset is strategically divided into an 80/20 ratio for training and validation sets while maintaining the directory structure. Various preprocessing and data augmentation techniques are applied to the input image samples. Subsequently, a Resnet9 model is employed to extract features. The developed model exhibits an impressive accuracy of 99.37 % and a minimal loss of 0.0018. This proposed model demonstrates its capability to achieve promising accuracy in distinguishing between healthy and diseased plants, showcasing the effectiveness of the developed approach.